A nomogram model integrating LI-RADS features based on MRI for predicting microvascular invasion in hepatocellular carcinoma following Milan criteria / 中华放射学杂志

ABSTRACT

Objective:To establish and verify a nomogram model based on MRI liver imaging reporting and data system (LI-RADS) features for predicting microvascular invasion (MVI) in hepatocellular carcinoma (HCC) following the Milan criteria.Methods:A retrospective analysis was conducted on data from 118 HCC patients (121 lesions) confirmed by pathology from June 2016 to June 2022 at the Third Affiliated Hospital of Soochow University. Forty-seven HCCs were diagnosed as MVI-positive and 74 HCCs as MVI-negative. The data was randomly divided into the training set (83 patients with 84 HCCs, including 31 MVI-positive and 53 MVI-negative HCCs) and the test set (35 patients with 37 HCCs, including 16 MVI-positive and 21 MVI-negative HCCs) using cross-validation method. HCC imaging features were evaluated based on LI-RADS (version 2018). In the training set, the χ 2 test was used to compare the differences in LI-RADS features between the MVI-positive group and the MVI-negative group. The logistic regression analysis was conducted to identify independent risk factors for predicting MVI-positive and to construct the nomogram model. The receiver operating characteristic (ROC) curves and decision curve analysis (DCA) were used to evaluate the performance and clinical benefits of the nomogram model in predicting MVI tumors. Results:There were statistically significant differences between the MVI-positive group and the MVI-negative group in terms of tumor size, tumor margin, mosaic architecture, and corona enhancement ( P<0.05). Multivariate logistic analysis results showed that HCC maximum diameter>3 cm (OR=1.427, 95%CI 1.314-12.227, P=0.009), nonsmooth tumor margin (OR=3.167, 95%CI 1.227-461.232, P=0.041), mosaic architecture (OR=1.769, 95%CI 1.812-61.434, P=0.022), and corona enhancement (OR=4.015, 95%CI 3.327-836.384, P=0.011) were independent risk factors for predicting MVI-positive tumors. Based on the independent predictors, the constructed nomogram model demonstrated an area under the ROC curve of 0.863 (95%CI 0.768-0.947) and 0.887 (95%CI 0.804-0.987) in the training and test sets for predicting MVI tumors, respectively. DCA showed that the curve of the nomogram model was consistently above the treat-all and treat-none strategies across all reasonable threshold probabilities in the training set, indicating that patients could obtain clinical benefits from the model. Conclusions:The preoperative nomogram model based on MRI LI-RADS features can effectively predict MVI in HCC following the Milan criteria, which could benefit the patients.